|本期目录/Table of Contents|

[1]范炤△,姚丽丽.基于功能磁共振成像的个体脑网络在阿尔兹海默症早期诊断中的应用*[J].生物医学工程研究,2021,02:126-130.
　FAN Zhao,YAO Lili.Application of functional magnetic resonance imaging based individual brain network in early diagnosis of Alzheimer′s disease[J].Journal of Biomedical Engineering Research,2021,02:126-130.
点击复制

基于功能磁共振成像的个体脑网络在阿尔兹海默症早期诊断中的应用*(PDF)

《生物医学工程研究》[ISSN:1006-6977/CN:61-1281/TN]

期数:
2021年02期
页码:
126-130
栏目:
出版日期:
2021-06-25

文章信息/Info

Title:
Application of functional magnetic resonance imaging based individual brain network in early diagnosis of Alzheimer′s disease
文章编号:
1672-6278 (2021)02-0126-05
作者:
范炤1△; 姚丽丽2
1.山西医科大学老年医学研究所,山西 太原 030001；2.山西医科大学基础医学院,山西 太原 030001
Author(s):
FAN Zhao1; YAO Lili2
1. Institute of Geriatrics, Shanxi Medical University, Tai yuan 030001，China；2. Department of Physiology, Basic Medical Sciences of Shanxi Medical University, Tai yuan 030001
关键词:
功能磁共振成像; 阿尔兹海默症; 脑网络; 机器学习; 核主成分分析
Keywords:
Functional magnetic resonance imaging;  Alzheimer′s disease;  Brain network;  Machine learning;  Kernel principal component analysis
分类号:
R318
DOI:
10.19529/j.cnki.1672-6278.2021.02.04
文献标识码:
A
摘要:
本研究利用功能磁共振成像（functional magnetic resonance imaging ,fMRI)构建个体脑网络以期能够对阿尔兹海默症（Alzheimer′s disease，AD）不同病程阶段进行分类，为临床诊断早期AD提供一种辅助手段。首先构建有向脑网络，将体素葡萄糖代谢平均率和脑网络连接，增加节点的度作为被研究对象对应图像的特征。然后采用Wrapper式特征选择法分别验证三种特征在核主成分分析（kernel principal component analysis，KPCA）和Adaboost两种机器学习算法下诊断AD的性能，将特征融合后以同样的方法进行验证。最后，对比分析了两种预测模型在AD不同病程中的分类性能，用十折交叉验证评估预测性能。结果显示，就单特征识别能力而言，平均葡萄糖代谢率对于AD的分类性能贡献最大，在两种算法下分别达到了93.21%和92.89%的准确率，多特征融合的分类性能最佳，准确率达94%以上，AUC值为0.97。两种算法模型对AD不同分类组的预测能力都不错，虽略有差异，但相比而言，KPCA算法表现更好。本研究可为计算机辅助AD早期诊断、及时干预提供参考依据。
Abstract:
To classify different stages of Alzheimer′s disease (AD),we used functional magnetic resonance imaging (fMRI) to construct individual brain networks, and provided an auxiliary means for early clinical diagnosis of AD. Firstly, a small block model was constructed to connect the average rate of voxel and glucose metabolism with the brain network; the degree of nodes was added as the feature of the object corresponding image.Then the Wrapper feature selection method was used to verify the performance of the three features in AD diagnosis under two machine learning algorithms of KPCA and Adaboost respectively. The features were fused and verified in the same way.Finally, the classification performance of the two prediction models in different course of AD was compared and analyzed, and the prediction performance was evaluated by ten fold cross validation. The results showed that, as far as the single feature recognition ability was concerned, the average glucose metabolism rate contributed the most to AD classification performance, with the accuracy of 93.21% and 92.89% respectively under the two algorithms. The classification performance of multi-feature fusion was the best with the accuracy of over 94% and the AUC value was 0.97. The predictive ability of the two algorithm models for different AD classification groups is good, slightly different, but compared with the KPCA algorithm, the performance is better.This study provides reference for the early diagnosis and timely intervention of computer-aided AD.

参考文献/References

备注/Memo

备注/Memo:
（收稿日期：2020-11-18）山西省科技厅国际合作项目(201803D421068)；山西省人社厅留学回国人员科技活动择优资助项目(619017)。△通信作者Email:fanzhao316@163.com
更新日期/Last Update: 2021-07-21